← All projects

Voice routing

A practical experiment in handling calls

Role Workflow design & testing
Focus Twilio, ElevenLabs, Telegram
Status Currently off
◖
Call routing Personal voice experiment
Illustrative
Incoming call
Route the call Contact list + voice assistant
Known contact Route onward Use contact-based rules
Other calls Voice assistant Recording and transcript
Review in Telegram ↗
Incoming call → Route → Review and adjust

A personal workflow built around interruptions

My first problem was the number of unwanted calls reaching my phone. I wanted known callers to have a direct route to me while unfamiliar calls could be understood before I decided whether to respond.

I put together a personal call-screening workflow with Twilio, ElevenLabs and Telegram. The contact list determined the first branch: recognised numbers went straight to me, while unknown callers spoke to an agent. The agent asked who was calling, the purpose of the call and how the person could be reached.

I then received a report, recording and transcript in Telegram and decided whether to call back. The workflow also shortened conversations identified as sales or spam. It changed the interruption into information I could review in my own time, while leaving me responsible for the response.

Reviewing real calls changed the instructions

When the assistant interpreted something incorrectly, I adjusted its prompt and followed up myself where needed. That feedback loop was part of the implementation. Connecting a number to a voice model was only the beginning; the instructions needed to work with the calls I actually received.

I observed fewer unwanted interruptions during use. The personal screening setup is now switched off. Its value as a project is the routing, intake and review process it let me build and examine.

Illustrative workflow

A personal call workflow

Incoming Call → contact-based routing → voice assistant where appropriate
Review Recording and transcript → Telegram → prompt adjustment
Status Personal experiment · currently off

A separate support-agent prototype for ÅlesHund

I also created a distinct ElevenLabs prototype for the shop. This was a customer-support experiment, separate from the personal phone filter. The build assembled a knowledge base from the public website, covering products, shipping, returns, care and sizing information, with Norwegian and English support.

The prototype was created in the ElevenLabs workspace with 38 website documents attached and structured data collection for the call. Its initial state used a bilingual menu. It could be tested through the web widget; the Twilio number and downstream webhook were not attached at that stage, so summaries remained in the platform dashboard.

That separation makes the prototype's purpose clear. It established a voice interface and knowledge context first, with the telephone and notification handoffs still to be connected. It was not presented as a fully launched phone-support operation.

A voice project needs clear routing, relevant context and a decision about what to collect and where to hand it over. I can help put those parts together, then test the channel and its review process.